Elliptic Curve Cryptography Speaker : Debdeep Mukhopadhyay Dept of Computer Sc and Engg IIT Madras

Outline of the Talk… Introduction to Elliptic Curves Elliptic Curve Cryptosystems (ECC) Implementation of ECC in Binary Fields

Introduction to Elliptic Curves

Lets start with a puzzle… What is the number of balls that may be piled as a square pyramid and also rearranged into a square array? Soln: Let x be the height of the pyramid… Thus, We also want this to be a square: Hence,

Graphical Representation Y axis X axis Curves of this nature are called ELLIPTIC CURVES

Method of Diophantus Uses a set of known points to produce new points (0,0) and (1,1) are two trivial solutions Equation of line through these points is y=x. Intersecting with the curve and rearranging terms: We know that 1 + 0 + x = 3/2 => x = ½ and y = ½ Using symmetry of the curve we also have (1/2,-1/2) as another solution

Diophantus’ Method Consider the line through (1/2,-1/2) and (1,1) => y=3x-2 Intersecting with the curve we have: Thus ½ + 1 + x = 51/2 or x = 24 and y=70 Thus if we have 4900 balls we may arrange them in either way

Elliptic curves in Cryptography Elliptic Curve (EC) systems as applied to cryptography were first proposed in 1985 independently by Neal Koblitz and Victor Miller. The discrete logarithm problem on elliptic curve groups is believed to be more difficult than the corresponding problem in (the multiplicative group of nonzero elements of) the underlying finite field.

Discrete Logarithms in Finite Fields F={1,2,3,…,p-1} Pick secret, random Y from F Pick secret, random X from F gx mod p gy mod p Alice Bob Compute k=(gy)x=gxy mod p Compute k=(gx)y=gxy mod p Eve has to compute gxy from gx and gy without knowing x and y… She faces the Discrete Logarithm Problem in finite fields

Elliptic Curve on a finite set of Integers Consider y2 = x3 + 2x + 3 (mod 5) x = 0  y2 = 3  no solution (mod 5) x = 1  y2 = 6 = 1  y = 1,4 (mod 5) x = 2  y2 = 15 = 0  y = 0 (mod 5) x = 3  y2 = 36 = 1  y = 1,4 (mod 5) x = 4  y2 = 75 = 0  y = 0 (mod 5) Then points on the elliptic curve are (1,1) (1,4) (2,0) (3,1) (3,4) (4,0) and the point at infinity:  Using the finite fields we can form an Elliptic Curve Group where we also have a DLP problem which is harder to solve…

Definition of Elliptic curves An elliptic curve over a field K is a nonsingular cubic curve in two variables, f(x,y) =0 with a rational point (which may be a point at infinity). The field K is usually taken to be the complex numbers, reals, rationals, algebraic extensions of rationals, p-adic numbers, or a finite field. Elliptic curves groups for cryptography are examined with the underlying fields of Fp (where p>3 is a prime) and F2m (a binary representation with 2m elements).

General form of a EC An elliptic curve is a plane curve defined by an equation of the form Examples

Weierstrass Equation A two variable equation F(x,y)=0, forms a curve in the plane. We are seeking geometric arithmetic methods to find solutions Generalized Weierstrass Equation of elliptic curves: Here, A, B, x and y all belong to a field of say rational numbers, complex numbers, finite fields (Fp) or Galois Fields (GF(2n)).

If Characteristic field is not 2: If Characteristics of field is neither 2 nor 3:

Points on the Elliptic Curve (EC) Elliptic Curve over field L It is useful to add the point at infinity The point is sitting at the top of the y-axis and any line is said to pass through the point when it is vertical It is both the top and at the bottom of the y-axis

The Abelian Group Given two points P,Q in E(Fp), there is a third point, denoted by P+Q on E(Fp), and the following relations hold for all P,Q,R in E(Fp) P + Q = Q + P (commutativity) (P + Q) + R = P + (Q + R) (associativity) P + O = O + P = P (existence of an identity element) there exists ( − P) such that − P + P = P + ( − P) = O (existence of inverses)

Elliptic Curve Picture y Consider elliptic curve E: y2 = x3 - x + 1 If P1 and P2 are on E, we can define P3 = P1 + P2 as shown in picture Addition is all we need P2 P1 x P3

Addition in Affine Co-ordinates y=m(x-x1)+y1 x y Let, P≠Q, y2=x3+Ax+B

Doubling of a point Let, P=Q What happens when P2=∞?

Why do we need the reflection? P2=O=∞ P1=P1+ O=P1 P1

Sum of two points Define for two points P (x1,y1) and Q (x2,y2) in the Elliptic curve Then P+Q is given by R(x3,y3) :

Point at infinity O P+P = 2P As a result of the above case P=O+P O is called the additive identity of the elliptic curve group. Hence all elliptic curves have an additive identity O.

Projective Co-ordinates Two-dimensional projective space over K is given by the equivalence classes of triples (x,y,z) with x,y z in K and at least one of x, y, z nonzero. Two triples (x1,y1,z1) and (x2,y2,z2) are said to be equivalent if there exists a non-zero element λ in K, st: (x1,y1,z1) = (λx2, λy2, λz2) The equivalence class depends only the ratios and hence is denoted by (x:y:z)

Projective Co-ordinates If z≠0, (x:y:z)=(x/z:y/z:1) What is z=0? We obtain the point at infinity. The two dimensional affine plane over K: There are advantages with projective co-ordinates from the implementation point of view

It is usual to assume the EC has no singular points Singularity For an elliptic curve y2=f(x), define F(x,y)=y2-F(x). A singularity of the EC is a pt (x0,y0) such that: It is usual to assume the EC has no singular points

If Characteristics of field is not 3: Hence condition for no singularity is 4A3+27B2≠0 Generally, EC curves have no singularity

Elliptic Curves in Characteristic 2 Generalized Equation: If a1 is not 0, this reduces to the form: If a1 is 0, the reduced form is: Note that the form cannot be:

Outline of the Talk… Introduction to Elliptic Curves Elliptic Curve Cryptosystems Implementation of ECC in Binary Fields

Elliptic Curve Cryptosystems (ECC)

Public-Key Cryptosystems Authentication: Only A can generate the encrypted message Secrecy: Only B can Decrypt the message Stallings Fig 9-4. Here see various components of public-key schemes used for both secrecy and authentication. Note that separate key pairs are used for each of these – receiver owns and creates secrecy keys, sender owns and creates authentication keys.

Public-Key Cryptography Stallings Fig 9-1.

Public-Key Cryptography Stallings Fig 9-1.

What Is Elliptic Curve Cryptography (ECC)? Elliptic curve cryptography [ECC] is a public-key cryptosystem just like RSA, Rabin, and El Gamal. Every user has a public and a private key. Public key is used for encryption/signature verification. Private key is used for decryption/signature generation. Elliptic curves are used as an extension to other current cryptosystems. Elliptic Curve Diffie-Hellman Key Exchange Elliptic Curve Digital Signature Algorithm

Using Elliptic Curves In Cryptography The central part of any cryptosystem involving elliptic curves is the elliptic group. All public-key cryptosystems have some underlying mathematical operation. RSA has exponentiation (raising the message or ciphertext to the public or private values) ECC has point multiplication (repeated addition of two points).

Generic Procedures of ECC Both parties agree to some publicly-known data items The elliptic curve equation values of a and b prime, p The elliptic group computed from the elliptic curve equation A base point, B, taken from the elliptic group Similar to the generator used in current cryptosystems Each user generates their public/private key pair Private Key = an integer, x, selected from the interval [1, p-1] Public Key = product, Q, of private key and base point (Q = x*B)

Example – Elliptic Curve Cryptosystem Analog to El Gamal Suppose Alice wants to send to Bob an encrypted message. Both agree on a base point, B. Alice and Bob create public/private keys. Alice Private Key = a Public Key = PA = a * B Bob Private Key = b Public Key = PB = b * B Alice takes plaintext message, M, and encodes it onto a point, PM, from the elliptic group

Example – Elliptic Curve Cryptosystem Analog to El Gamal Alice chooses another random integer, k from the interval [1, p-1] The ciphertext is a pair of points PC = [ (kB), (PM + kPB) ] To decrypt, Bob computes the product of the first point from PC and his private key, b b * (kB) Bob then takes this product and subtracts it from the second point from PC (PM + kPB) – [b(kB)] = PM + k(bB) – b(kB) = PM Bob then decodes PM to get the message, M.

Example – Compare to El Gamal The ciphertext is a pair of points PC = [ (kB), (PM + kPB) ] The ciphertext in El Gamal is also a pair. C = (gk mod p, mPBk mod p) -------------------------------------------------------------------------- Bob then takes this product and subtracts it from the second point from PC (PM + kPB) – [b(kB)] = PM + k(bB) – b(kB) = PM In El Gamal, Bob takes the quotient of the second value and the first value raised to Bob’s private value m = mPBk / (gk)b = mgk*b / gk*b = m

Diffie-Hellman (DH) Key Exchange

ECC Diffie-Hellman Public: Elliptic curve and point B=(x,y) on curve Secret: Alice’s a and Bob’s b a(x,y) b(x,y) Alice, A Bob, B Alice computes a(b(x,y)) Bob computes b(a(x,y)) These are the same since ab = ba

Example – Elliptic Curve Diffie-Hellman Exchange Alice and Bob want to agree on a shared key. Alice and Bob compute their public and private keys. Alice Private Key = a Public Key = PA = a * B Bob Private Key = b Public Key = PB = b * B Alice and Bob send each other their public keys. Both take the product of their private key and the other user’s public key. Alice  KAB = a(bB) Bob  KAB = b(aB) Shared Secret Key = KAB = abB

Why use ECC? How do we analyze Cryptosystems? How difficult is the underlying problem that it is based upon RSA – Integer Factorization DH – Discrete Logarithms ECC - Elliptic Curve Discrete Logarithm problem How do we measure difficulty? We examine the algorithms used to solve these problems

Security of ECC To protect a 128 bit AES key it would take a: RSA Key Size: 3072 bits ECC Key Size: 256 bits How do we strengthen RSA? Increase the key length Impractical?

Applications of ECC Many devices are small and have limited storage and computational power Where can we apply ECC? Wireless communication devices Smart cards Web servers that need to handle many encryption sessions Any application where security is needed but lacks the power, storage and computational power that is necessary for our current cryptosystems

Benefits of ECC Same benefits of the other cryptosystems: confidentiality, integrity, authentication and non-repudiation but… Shorter key lengths Encryption, Decryption and Signature Verification speed up Storage and bandwidth savings

Summary of ECC “Hard problem” analogous to discrete log Q=kP, where Q,P belong to a prime curve given k,P  “easy” to compute Q given Q,P  “hard” to find k known as the elliptic curve logarithm problem k must be large enough ECC security relies on elliptic curve logarithm problem compared to factoring, can use much smaller key sizes than with RSA etc for similar security ECC offers significant computational advantages

Outline of the Talk… Introduction to Elliptic Curves Elliptic Curve Cryptosystems Implementation of ECC in Binary Fields

Implementation of ECC in Binary Fields

Sub-Topics Scalar Multiplication: LSB first vs MSB first Montgomery Technique of Scalar Multiplication Fast Scalar Multiplication without pre-computation. Lopez and Dahab Projective Transformation to Reduce Inverters Mixed Coordinates Parallelization Techniques Half and Add Technique for Scalar Multiplication

ECC operations: Hierarchy Level 0 Level 1 Level 2 Level 3

Scalar Multiplication: MSB first Require k=(km-1,km-2,…,k0)2, km=1 Compute Q=kP Q=P For i=m-2 to 0 Q=2Q If ki=1 then Q=Q+P End if End for Return Q Sequential Algorithm Requires m point doublings and (m-1)/2 point additions on the average

Example Compute 7P: Compute 6P: 7=(111)2 7P=2(2(P)+P)+P=> 2 iterations are required Principle: First double and then add (accumulate) Compute 6P: 6=(110)2 6P=2(2(P)+P)

Scalar Multiplication: LSB first Require k=(km-1,km-2,…,k0)2, km=1 Compute Q=kP Q=0, R=P For i=0 to m-1 If ki=1 then Q=Q+R End if R=2R End for Return Q Can Parallelize… What you are doubling and what you are accumulating are different… On the average m/2 point Additions and m/2 point doublings

Example Compute 7P, 7=(111)2, Q=0, R=P Compute 6P, 6=(110)2, Q=0, R=P Q=Q+R=0+P=P, R=2R=2P Q=P+2P=3P, R=4P Q=7P, R=8P Compute 6P, 6=(110)2, Q=0, R=P Q=0, R=2R=2P Q=0+2P=2P, R=4P Q=2P+4P=6P, R=8P

Compute 31P… Q=2P Q=P, R=2P Q=3P Q=3P, R=4P Q=6P Q=7P, R=8P Q=7P 31=(11111)2 MSB First LSB First Q=2P Q=3P Q=6P Q=7P Q=14P Q=15P Q=30P Q=31P Q=P, R=2P Q=3P, R=4P Q=7P, R=8P Q=15P, R=16P Q=31P, R=32P

Weierstrass Point Addition Let, P=(x1,y1) be a point on the curve. -P=(x1,x1+y1) Let, R=P+Q=(x3,y3) Point addition and doubling each require 1 inversion & 2 multiplications 2. We neglect the costs of squaring and addition 3. Montgomery noticed that the x-coordinate of 2P does not depend on the y-coordinate of P

Montgomery’s method to perform scalar multiplication Input: k>0, P Output: Q=kP Set k<-(kl-1,…,k1,k0)2 Set P1=P, P2=2P For i from l-2 to 0 If ki=1, Set P1=P1+P2, P2=2P2 else Set P2=P2+P1, P1=2P1 Return Q=P1 Invariant Property: P=P2-P1 Question: How to implement the Operation efficiently?

Example Compute 7P 7=(111)2 Initialization: P1=P; P2=2P Steps: 7=(110)2 Initialization: P1=P; P2=2P Steps: P1=3P, P2=4P P2=7P, P1=6P

Fast Multiplication on EC without pre-computation

Result-1 Let P1 = (x1,y1) and P2=(x2,y2) be elliptic points. Then the x-coordinate of P1+P2, x3 can be computed as: Hint: Remember that the field has a characteristic 2 and that P1 and P2 are points on the curve

Result-2 Let P=(x,y), P1 = (x1,y1) and P2=(x2,y2) be elliptic points. Let P=P2-P1 be an invariant. Then the x-coordinate of P1+P2, x3 can be computed in terms of the x-coordinates as:

Result-3 Let P=(x,y), P1=(x1,y1) and P2=(x2,y2) be elliptic points. Assume that P2-P1=P and x is not 0. Then the y-coordinates of P1 can be expressed in terms of P, and the x-coordinates of P1 and P2 as follows:

Final Algorithm #INV:2(l-2)+1; #MULT: 2(l-2)+4 #ADD: 4(l-2)+6 Input: k>0, P=(x,y) Output: Q=kP If k=0 or x=0 then output(0,0) Set k = (kl-1,kl-2,…,k0)2 Set x1=x, x2=x2+b/x2 For i from l-2 to 0 Set t=x1/(x1+x2) If ki=1, x1=x+t2+t, x2=x22+b/x22 else x1=x12+b/x12, x2=x+t2+t r1=x1+x, r2=x2+x y1=r1(r1r2+x2+y)/x+y Return Q=(x1,y1) #INV:2(l-2)+1; #MULT: 2(l-2)+4 #ADD: 4(l-2)+6 #SQR: 2(l-2)+2

How to reduce inversions? In affine coordinates Inverses are very expensive For each inversion requires around 7 multipliers (in hardware designs) Lopez Dahab Projective coordinates: (X,Y,Z), Z≠0, maps to (X/Z,Y/Z2) Motivation is to replace inversions by the multiplication operations and then perform one inversion at the end (to obtain back the affine coordinates)

Doubling Remember: In Projective Coordinates: 2 inverses 1 general field multiplication 4 additions 2 squarings Remember: In Projective Coordinates: 0 inverses 4 general field multiplications 3 additions 5 squarings

Montgomery Algorithm Input: k>0, P=(x,y) Output: Q=kP Set k<-(kl-1,…,k1,k0)2 Set X1=x, Z1=1; X2=x4+b, Z2=x2 For i from l-2 to 0 If ki =1, Madd(X1,Z1,X2,Z2), Mdouble(X2,Z2) else Madd(X2,Z2,X1,Z1), Mdouble(X1,Z1) Return Q=(Mxy(X1,Y1,X2,Y2))

Mxy: Projective to Affine Requires 10 multiplications and one inverse operation

Final Comparison Affine Coordinates Inv: 2logk + 1 Mult: 2logk + 4 Add: 4logk + 6 Sqr: 2logk + 2 Projective Coordinates Inv: 1 Mult: 6logk + 10 Add: 3logk + 7 Sqr: 5logk + 3 Hence, final decision depends upon the I:M ratio of the finite field operators

Addition in Mixed Coordinates Theorem: Let P1=(X1/Z1,Y1/Z12) and P2=(X2/Z2,Y2/Z22) be two points on the curve. If Z1=1, then P1+P2=(X3/Z3,Y3/Z32) st. Number of multiplications are further reduced. Squaring is increased a bit, but they are cheap in GF(2n) Improvement by 10 % if a≠0, otherwise 12 %...

Parallel Strategies for Scalar Point Multiplication Point Doubling Cycle 1: T=X12, M=cZ12, Z2=T.Z12 Cycle 1a: X2=T2+M2 Point Addition Cycle 1: t1=(X1.Z2); t2=(Z1.X2) Cycle 1a: M=(t1+t2), Z1=M2 Cycle 2: N=t1.t2, M=xZ1 Cycle 2a: X1=M+N 1 multiplier 2 multipliers We assume that squarings and multiplications with constants can be performed without multipliers…

Parallelizing Montgomery Algorithm Input: k>0, P=(x,y) Output: Q=kP Set k<-(kl-1,…,k1,k0)2 Set X1=x, Z1=1; X2=x4+b, Z2=x2 For i from l-2 to 0 If ki =1, 5a) Madd(X1,Z1,X2,Z2), Mdouble(X2,Z2) else 5b) Madd(X2,Z2,X1,Z1), Mdouble(X1,Z1) Return Q=(Mxy(X1,Y1,X2,Y2))

Looking back at our Design Hierarchy Level 0 Level 1 Level 2 Level 3

Parallelizing Strategies Parallelize level 1: If we allocate one multiplier to each of Madd and Mdouble, then we can parallelize steps 5a and 5b. Thus 4 clock cycles are required for each iteration. Total time is nearly 4l. Parallelize level 2: If we can parallelize the underlying Madd and Mdouble, then we cannot parallelize level 1, if we have constraint of 2 multipliers. So, we have a sequential step 5a and 5b. Total time is 3l.

Parallelizing Strategies Parallelize both the levels: Total time is 2l clock cycles. Require 3 multipliers. Thus Montgomery algorithm is highly parallelizable Helpful in high performance designs (low power, high thoughput etc)

Point Halving In 1999 Scroeppel and Knudsen proposed further speed up Idea is to replace point doubling by halving Point Halving is three times as fast than doubling The scalar k, has to be expressed in the negative powers of 2

Computing the Half Problem: Let E be the Elliptic Curve, defined by the equation: Let Q=(u,v)=2P Compute P=(x,y) Remember :

Halving (contd.) Thus, we have to solve the above equations Square Root Solving Quadratics Thus, we have to solve the above equations λ-representation: (x, λx)

Trace of a point Define: Properties of Trace: Tr(c)=Tr(c2)=Tr(c)2, Tr(c) can be 0 or 1 Tr(c+d)=Tr(c)+Tr(d) NIST Curves : Tr(a)=1 If x,y belongs to the Elliptic Curve, Tr(x)=Tr(a)

Computing λ The roots of are λ1= λ or λ+1 Theorem:

Halving Algorithm Input: (u,v) , Output: (x,y) Solve for λ. Let the root be Compute If Tr(t)=0, then λP= , x=(t+u)1/2 else λP= +1,x=(t)1/2 4. Return (x,λP)

Implementation of Trace Can be evaluated in O(1) time Example: GF(2163), with reduction polynomial p(x)=x163+x7+x6+x3+1, Tr(xi)=1, iff i=0 or 159. Thus, the implementation is only one xor gate to add the 0th and the 159th bits of the register storing C.

Solving a Quadratic over GF(2m) Solve x2+x=c+Tr(c), c is an element of GF(2m) Define Half Trace: H(C) gives a root for the quadratic equation. A simple method to find H(C) requires storage for m elements and m/2 field additions on an average

Obtaining Square Root Field squaring in binary field is linear Hence squaring can be rephrased as: C=MA=A2 We require to compute D st. D2=A Let, D=M-1A=> A=MD D2=MD (as M is the squaring matrix) =M(M-1A)=A Hence, D=(A)1/2

An Example

Half and Add Algorithm Input: 0<k<n, P=(x,y) Output: Q=kP Compute: , k1=(2t-1k)mod n Q=O for i=0 to m-1 do Q=[1/2]Q If, k1i=1, then Q=Q+P return Q No method is currently known to perform point halving in projective Coordinates. Keep Q in affine coordinates and P in Projective Coordinates. Then step 5.2 is a mixed operation, giving further efficiency.

Key References Papers: J. Lopez and R. Dahab, “Fast Multiplication on Elliptic Curves over GF(2m) without pre-computation”, CHES 1999 K. Fong etal, “Field Inversion and Point Halving Revisited”, IEEE Trans on Comp, 2004 G. Orlando and C. Paar, “A High Performance Reconfigurable Elliptic Curve Processor for GF(2m)”, CHES 2000 N. A. Saqib etal, “A Parallel Architecture for Fast Computation of Elliptic Curve Scalar Multiplication over GF(2m)”, Elsevier Journal of Microprocessors and Microsystems, 2004 Sabiel Mercurio etal, “ An FPGA Arithmetic Logic Unit for Computing Scalar Multiplication using the Half-and-Add Method”, IEEE ReConfig 2005

Key References Books: Elliptic Curves: Number Theory and Cryptography, by Lawrence C. Washington Guide to Elliptic Curve Cryptography, Alfred J. Menezes Guide to Elliptic Curve Cryptography, Darrel R. Hankerson, A. Menezes and A. Vanstone http://cr.yp.to/ecdh.html ( Daniel Bernstein)

Thank You